Research And Implementation Of Deep Sea Image Preprocessing And Online Stitching Method

Deep sea exploration has broad development prospects.The research,development and utilization of deep-sea resources have become an important demand for national sustainable development.As an important technical means of deep-sea exploration,image mosaic provides detailed visual information of large scene seabed environment.Therefore,seabed image mosaic technology has important research value.In the problem of seabed image mosaic,due to the imaging characteristics of underwater environment,seabed images often have problems such as light attenuation,uneven illumination and light scattering,which leads to the performance of classical feature detection and matching algorithms is greatly reduced when applied to seabed images,which seriously affects the automatic realization of mosaic tasks.Secondly,in the application of robot based seabed exploration,the traditional image mosaic is generally offline.This kind of method can not understand and master the large-scale scene information of seabed in real time,therefore control the motion of robot more effectively.In view of the above two problems,this thesis makes relevant research,and the main work is as follows:(1)Research on seabed image preprocessing method.Aiming at the main factors affecting image matching,based on a large number of theoretical analysis and experiments,this thesis proposes a set of seabed image preprocessing methods to improve the performance of feature matching,including two image characteristics(resolution and color channel)and three image processing methods(uneven illumination compensation,linear contrast stretching and Gaussian filter smoothing).SIFT matching algorithm is used to study the influence of the above preprocessing methods on the number of features and the performance of feature matching.The experimental results show that using the image with 1/2 resolution for feature detection and matching can reduce the calculation time and ensure more interior point ratio than the original image.The matching performance of blue channel image is better than that of red channel,green channel and gray image.Illumination compensation improves the phenomenon of uneven illumination in the scene,adds a small number of feature points,and ensures the consistency of subsequent image fusion.The linear contrast enhancement method is more reasonable than the nonlinear enhancement method,which increases the number of feature points matching.Gaussian filter smoothing effectively reduces the inverse influence of the image itself and the noise added by contrast enhancement on feature matching.The combination of the above methods is applied to feature matching.Compared with the image without preprocessing,it not only improves the visual effect of image to a certain extent,but also effectively improves the feature matching performance between images.The number of successfully matched features is increased by about 1.57 times.(2)Research on online stitching method of seabed image based on recursive least square method.In the traditional offline stitching task based on linear motion model,the global alignment is generally realized by the global linear least square method.However,for the online stitching problem discussed in this thesis,with the continuous increase of new data,the coefficient matrix in the least squares method will become larger and larger,and the corresponding solution time will be longer and longer,which can not meet the needs of practical applications.To solve this problem,this thesis presents an online seabed image mosaic method based on recursive least squares to solve the global alignment model.Using this method,every time a new image is added,the global alignment model can be quickly updated by using a small amount of old and new data.This is also the key problem to realize online stitching.The implementation results verify the feasibility and efficiency of this method.(3)Design and development of online stitching prototype system for seabed images.According to the needs of deep-sea exploration,based on the above work,this thesis designs and develops an interactive online stitching prototype system.The system integrates the seabed image preprocessing steps proposed in this thesis with the online stitching algorithm,including four modules: image input,image preprocessing,image online stitching and panorama output.The test results show that the system runs correctly and has certain practicability.